Inteligência artificial na modelagem do equilíbrio líquido-vapor de misturas binárias: estado da arte e perspectivas futuras
Arquivos
Data
2021-08-04
Autores
Ottaiano, Gabriel Yoshiaki 
Orientadores
Martins, Tiago Dias
Tipo
Trabalho de conclusão de curso
Título da Revista
ISSN da Revista
Título de Volume
Resumo
O Equilíbrio Líquido-Vapor (ELV) está presente em diversas áreas das indústrias químicas, principalmente no processo de separação de compostos. O estudo de sua modelagem se torna essencial para o projeto de processos no que tange o dimensionamento e controle de equipamentos. As Inteligências Artificiais vêm sendo desenvolvidas e utilizadas cada vez mais, em variadas áreas, tanto na pesquisa quanto em aplicações no dia-a-dia da sociedade. Conforme ampliou-se o uso das inteligências artificiais, foram desenvolvidos trabalhos que visavam a utilização de modelos para a predição e cálculo do equilíbrio líquido-vapor. Devido à ausência de um panorama geral nesta área, o objetivo deste trabalho foi a realização de uma revisão da literatura sobre a modelagem do equilíbrio líquido-vapor de misturas binárias via inteligência artificial, no que concerne os diversos aspectos como a escolha das variáveis de entrada e saída dos modelos, métodos de otimização, classificação dos compostos, etc.. Com isso, esperou-se criar um trabalho que servirá como base de conhecimento para pesquisas futuras na área. As Redes Neurais Artificiais foram o modelo mais utilizado dentre os trabalhos encontrados, com resultados equiparados com modelos já consagrados como as Equações de Estado. Diferentes abordagens foram utilizadas, na maioria, os erros alcançados foram abaixo dos obtidos pelas Equações de Estado. As RNAs estão cada vez mais sendo desenvolvidas e aplicadas por terem alto poder de aprendizado. Porém há desvantagens, como a sensação de trabalhar com uma caixa-preta devido a difícil tradução de seus parâmetros numéricos a fenômenos reais. Outros métodos de inteligência artificial como programação genética, Support Vector Machine e Redes Neuro-Fuzzy também foram estudas com resultados promissores para a predição do ELV. Este trabalho apresentou padrões recorrentes na modelagem do ELV binário como a utilização de RNAs com apenas uma camada intermediária e o uso do algoritmo Levenberg-Marquardt, presente na maioria dos trabalhos. Além disso mostrou e comparou a capacidade de modelos de inteligência artificial de calcular o ELV de maneira tão ou mais eficiente que equações fenomenológicas.
Vapor-Liquid Equilibrium is present in several areas of chemical industries, especially in the process of separation of compounds. The study of its modeling becomes essential for the design of processes in terms of equipment design and control. Artificial Intelligences have been increasingly developed and used in several areas, both in research and in daily applications in the society. As the use of artificial intelligences increased, researches were developed that aimed at the use of models for the prediction and calculation of the vapor-liquid equilibrium. Due to the absence of a review work in this area, the objective of this research was to conduct a literature review on the modeling of the vapor-liquid equilibrium of binary mixtures via artificial intelligence, regarding the various aspects such as the choice of input and output variables, optimization methods, classification of compounds, etc., to create work that will serve as a knowledge base for future research in the area. Artificial Neural Networks (ANNs) were the most used model among the articles found, with results comparable to the models already established, such as the Equations of State. Different approaches were used, in most, the errors achieved were below those obtained by the Equations of State. The ANNs are increasingly being developed and applied as a powerful tool because of their high learning and adaptive flexibility. However, there are some disadvantages, such as working with a black box like tool due to the difficult translation of its numerical parameters to the real phenomena. Other artificial intelligence methods such as genetic programming, Support Vector Machine and Neuro-Fuzzy Networks have also been studied with promising results for VLE prediction. This work presented recurring patterns in binary VLE modeling such as the use of ANNs with only one intermediate layer and the use of the Levenberg-Marquardt algorithm, present in most works. Furthermore, was confirmed the ability of artificial intelligence models to calculate the VLE as efficiently or more than phenomenological equations.
Vapor-Liquid Equilibrium is present in several areas of chemical industries, especially in the process of separation of compounds. The study of its modeling becomes essential for the design of processes in terms of equipment design and control. Artificial Intelligences have been increasingly developed and used in several areas, both in research and in daily applications in the society. As the use of artificial intelligences increased, researches were developed that aimed at the use of models for the prediction and calculation of the vapor-liquid equilibrium. Due to the absence of a review work in this area, the objective of this research was to conduct a literature review on the modeling of the vapor-liquid equilibrium of binary mixtures via artificial intelligence, regarding the various aspects such as the choice of input and output variables, optimization methods, classification of compounds, etc., to create work that will serve as a knowledge base for future research in the area. Artificial Neural Networks (ANNs) were the most used model among the articles found, with results comparable to the models already established, such as the Equations of State. Different approaches were used, in most, the errors achieved were below those obtained by the Equations of State. The ANNs are increasingly being developed and applied as a powerful tool because of their high learning and adaptive flexibility. However, there are some disadvantages, such as working with a black box like tool due to the difficult translation of its numerical parameters to the real phenomena. Other artificial intelligence methods such as genetic programming, Support Vector Machine and Neuro-Fuzzy Networks have also been studied with promising results for VLE prediction. This work presented recurring patterns in binary VLE modeling such as the use of ANNs with only one intermediate layer and the use of the Levenberg-Marquardt algorithm, present in most works. Furthermore, was confirmed the ability of artificial intelligence models to calculate the VLE as efficiently or more than phenomenological equations.